from datasets import load_dataset
from sentence_transformers import SentenceTransformer
import torch

device = torch.device("cuda")
embedder = SentenceTransformer('all-mpnet-base-v2')

dataset = load_dataset("csv", data_dir="./raw", sep="|", split="train", column_names=['id', 'type', 'language', 'title'])
dataset.save_to_disk("./datasets")

embeddings = embedder.encode(dataset['title'], convert_to_tensor=True, device=device)
torch.save(embeddings, "./embeddings/corpus.pt")